瘤内及瘤周动态增强MRI影像组学特征鉴别乳腺原位癌与浸润性癌的可行性研究

A feasibility study of classification between breast carcinoma in situ and invasive carcinoma using intratumoral and peritumoral radiomics based on dynamic contrast-enhanced MRI

目的探讨基于乳腺癌瘤内及瘤周区域动态增强MRI影像组学特征构建的预测模型鉴别乳腺原位癌与浸润性癌的可行性。方法回顾性分析北京大学第一医院2013年1至12月经粗针穿刺活体组织检查或手术切除组织病理学检查确诊为浸润性乳腺癌及2013年1月至2015年12月经手术切除组织病理学检查确诊为乳腺原位癌的连续病例。所有患者均有术前完整乳腺MRI图像。共入组251例患者(251个病灶),均为女性,年龄23~82岁,中位年龄53岁,其中乳腺原位癌43例,乳腺浸润性癌208例。所有病例按照7∶3比例随机分为训练集(176例)与测试集(75例)。在训练集中结合动态增强(DCE)蒙片及早期强化图像,对乳腺肿瘤区域及瘤周区域进行半自动分割,然后提取、筛选影像组学特征并建立预测模型。测试集进行模型效能测试。采用受试者操作特征(ROC)曲线及曲线下面积(AUC)分析训练集及测试集中各模型的预测效能。结果利用瘤内、瘤周、瘤内联合瘤周影像组学特征建立的预测模型效能均较好,在测试集中3个预测模型鉴别乳腺原位癌与浸润性癌的AUC分别为0.865、0.896、0.922,两两比较差异无统计学意义(P>0.05),灵敏度分别为77.4%、87.1%、83.9%,特异度分别为92.3%、84.6%、100%,准确度分别为80.0%、85.3%、86.7%。结论利用乳腺癌瘤内及瘤周动态增强MRI的影像组学特征鉴别乳腺原位癌与浸润癌具有潜在可行性。

Objective To explore the feasibility of classification between carcinoma in situ and invasive carcinoma of breast using intratumoral and peritumoral radiomics based on breast dynamic contrast-enhanced(DCE)MRI.Methods The retrospective study included consecutive invasive breast carcinoma pathological diagnosed by core needle biopsy or surgery from January 2013 to December 2013 and carcinoma in situ of breast diagnosed by surgery from January 2013 to December 2015 in Peking University First Hospital.All patients had pretreatment breast MRI images.A total of 251 cases(251 lesions)were included,with 208 invasive breast carcinoma and 43 carcinoma in situ of breast.They were all females and median age was 53(23-82)years old.Patients were randomly divided into the training(n=176)and testing dataset(n=75)in a 7∶3 ratio.In the training dataset,combined with DCE mask and early enhancement images,intratumoral and peritumoral area were semi-automatic segmentation,and radiomics features were extracted and dimension reduction,finally a prediction model was established.Model performance was tested in the testing dataset.Receiver operating characteristic(ROC)curve and area under curve(AUC)were used to analyze the model prediction performance.Results The prediction models established by intratumoral,peritumoral and intratumoral combined with peritumoral radiomics had good performance.The AUC of intratumoral,peritumoral and intratumoral combined with peritumoral radiomics prediction models in differentiating breast carcinoma in situ and invasive carcinoma were 0.865,0.896 and 0.922 in the testing dataset,there was no significant difference in pairwise comparisons(P>0.05).The sensitivity of intratumoral,peritumoral and intratumoral combined with peritumoral radiomics prediction models were 77.4%,87.1%,83.9%,the specificity were 92.3%,84.6%,100%,and the accuracy were 80.0%,85.3%,86.7%.Conclusion It is potential feasible for classification between carcinoma in situ and invasive carcinoma of breast using intratumoral and peritumoral radiomics based on breast DCE MRI.